Merge branch 'develop-team-change-reset' into asymm-envs

.pre-commit-config.yaml

     hooks:
     -   id: markdown-link-check
         name: markdown-link-check
-        # markdown-link-check doesn't support multiple files on the commandline, so this hacks around that.
-        # Note that you must install the package separately via npm. For example:
-        #  brew install npm; npm install -g markdown-link-check
-        entry: bash -c 'for i in "$@"; do markdown-link-check -c markdown-link-check.fast.json "$i"; done' --
-        language: system
+        entry: utils/run_markdown_link_check.py
+        language: script
         types: [markdown]
         # Don't check localized files since their target might not be localized.
         exclude: ".*localized.*"
         name: markdown-link-check-full
-        entry: bash -c 'for i in "$@"; do markdown-link-check -c markdown-link-check.full.json "$i"; done' --
-        language: system
+        entry: utils/run_markdown_link_check.py
+        language: script
+        # Don't check localized files since their target might not be localized.
+        # Only run manually, e.g. pre-commit run --hook-stage manual markdown-link-check-full
+        args: [--check-remote]
     -   id: validate-versions
         name: validate library versions
         language: script

com.unity.ml-agents/Tests/Editor/ModelRunnerTest.cs

     [TestFixture]
     public class ModelRunnerTest
     {
-        const string k_continuous2vis8vec2actionPath = "Packages/com.unity.ml-agents/Tests/Editor/Resources/continuous2vis8vec2action.nn";
-        const string k_discrete1vis0vec_2_3action_recurrModelPath = "Packages/com.unity.ml-agents/Tests/Editor/Resources/discrete1vis0vec_2_3action_recurr.nn";
+        const string k_continuous2vis8vec2actionPath = "Packages/com.unity.ml-agents/Tests/Editor/TestModels/continuous2vis8vec2action.nn";
+        const string k_discrete1vis0vec_2_3action_recurrModelPath = "Packages/com.unity.ml-agents/Tests/Editor/TestModels/discrete1vis0vec_2_3action_recurr.nn";
         NNModel continuous2vis8vec2actionModel;
         NNModel discrete1vis0vec_2_3action_recurrModel;
         Test3DSensorComponent sensor_21_20_3;

com.unity.ml-agents/Tests/Editor/ParameterLoaderTest.cs

     [TestFixture]
     public class ParameterLoaderTest
     {
-        const string k_continuous2vis8vec2actionPath = "Packages/com.unity.ml-agents/Tests/Editor/Resources/continuous2vis8vec2action.nn";
-        const string k_discrete1vis0vec_2_3action_recurrModelPath = "Packages/com.unity.ml-agents/Tests/Editor/Resources/discrete1vis0vec_2_3action_recurr.nn";
+        const string k_continuous2vis8vec2actionPath = "Packages/com.unity.ml-agents/Tests/Editor/TestModels/continuous2vis8vec2action.nn";
+        const string k_discrete1vis0vec_2_3action_recurrModelPath = "Packages/com.unity.ml-agents/Tests/Editor/TestModels/discrete1vis0vec_2_3action_recurr.nn";
         NNModel continuous2vis8vec2actionModel;
         NNModel discrete1vis0vec_2_3action_recurrModel;
         Test3DSensorComponent sensor_21_20_3;

docs/Installation.md

 just cloned. You can add the `com.unity.ml-agents` package to
 your project by navigating to the menu `Window`  -> `Package Manager`. In the package manager
 window click on the `+` button. Select `Add package from disk...` and navigate into the
-`com.unity.ml-agents` folder and select the `package.json` folder.
+`com.unity.ml-agents` folder and select the `package.json` file.
 
 **NOTE:** In Unity 2018.4 it's on the bottom right of the packages list, and in Unity 2019.3 it's
 on the top left of the packages list.

docs/Training-Curriculum-Learning.md

 
 ## An Instructional Example
 
-*[**Note**: The example provided below is for instructional purposes, and was based on an early version of the [Wall Jump example environment](Example-Environments.md). As such, it is not possible to directly replicate the results here using that environment.]*
+*[**Note**: The example provided below is for instructional purposes, and was based on an early version of the [Wall Jump example environment](Learning-Environment-Examples.md).
+As such, it is not possible to directly replicate the results here using that environment.]*
 
 Imagine a task in which an agent needs to scale a wall to arrive at a goal. The
 starting point when training an agent to accomplish this task will be a random
 
 You can then keep track of the current lessons and progresses via TensorBoard.
 
-__Note__: If you are resuming a training session that uses curriculum, please pass the number of the last-reached lesson using the `--lesson` flag when running `mlagents-learn`.
+__Note__: If you are resuming a training session that uses curriculum, please pass the number of the last-reached lesson using the `--lesson` flag when running `mlagents-learn`.

docs/Training-on-Microsoft-Azure.md

 
 A pre-configured virtual machine image is available in the Azure Marketplace and
 is nearly completely ready for training. You can start by deploying the
-[Data Science Virtual Machine for Linux (Ubuntu)](https://azuremarketplace.microsoft.com/en-us/marketplace/apps/microsoft-dsvm.linux-data-science-vm-ubuntu)
+[Data Science Virtual Machine for Linux (Ubuntu)](https://azuremarketplace.microsoft.com/en-us/marketplace/apps/microsoft-dsvm.ubuntu-1804)
 into your Azure subscription.
 
 Note that, if you choose to deploy the image to an

ml-agents/mlagents/trainers/agent_processor.py

             if not terminated:
                 self.episode_steps[global_id] += 1
 
-            # if the trajectory is too long, we truncate it
+            # Add a trajectory segment to the buffer if terminal or the length has reached the time horizon
             if (
                 len(self.experience_buffers[global_id]) >= self.max_trajectory_length
                 or terminated

ml-agents/mlagents/trainers/ghost/controller.py

         self._learning_team: int = -1
         # Dict from team id to GhostTrainer for ELO calculation
         self._ghost_trainers: Dict[int, GhostTrainer] = {}
+        # Signals to the trainer control to perform a hard reset
+        self._reset = False
 
     @property
     def get_learning_team(self) -> int:
         """
         return self._learning_team
+
+    @property
+    def reset(self) -> bool:
+        """
+        Whether or not team change occurred. Causes full reset in trainer_controller
+        :return: The truth value of the team changing
+        """
+        change_team = self._reset
+        if self._reset:
+            self._reset = False
+        return change_team
 
     def subscribe_team_id(self, team_id: int, trainer: GhostTrainer) -> None:
         """
         logger.debug(
             "Learning team {} swapped on step {}".format(self._learning_team, step)
         )
+        self._reset = True
 
     # Adapted from https://github.com/Unity-Technologies/ml-agents/pull/1975 and
     # https://metinmediamath.wordpress.com/2013/11/27/how-to-calculate-the-elo-rating-including-example/

ml-agents/mlagents/trainers/learn.py

 from mlagents import tf_utils
 from mlagents.trainers.trainer_controller import TrainerController
 from mlagents.trainers.meta_curriculum import MetaCurriculum
+from mlagents.trainers.ghost.controller import GhostController
 from mlagents.trainers.trainer_util import (
     load_config,
     TrainerFactory,
         sampler_manager, resampling_interval = create_sampler_manager(
             options.sampler_config, run_seed
         )
+        ghost_controller = GhostController()
+
         trainer_factory = TrainerFactory(
             options.trainer_config,
             summaries_dir,
             not options.inference,
             options.resume,
             run_seed,
+            ghost_controller,
             maybe_init_path,
             maybe_meta_curriculum,
             options.multi_gpu,
             run_seed,
             sampler_manager,
             resampling_interval,
+            ghost_controller,
         )
 
     # Begin training

ml-agents/mlagents/trainers/trainer_controller.py

 from mlagents.trainers.trainer_util import TrainerFactory
 from mlagents.trainers.behavior_id_utils import BehaviorIdentifiers
 from mlagents.trainers.agent_processor import AgentManager
+from mlagents.trainers.ghost.controller import GhostController
 
 
 class TrainerController(object):
         training_seed: int,
         sampler_manager: SamplerManager,
         resampling_interval: Optional[int],
+        ghost_controller: GhostController,
     ):
         """
         :param model_path: Path to save the model.
         self.meta_curriculum = meta_curriculum
         self.sampler_manager = sampler_manager
         self.resampling_interval = resampling_interval
+        self.ghost_controller = ghost_controller
 
         self.trainer_threads: List[threading.Thread] = []
         self.kill_trainers = False
             and (self.resampling_interval)
             and (steps % self.resampling_interval == 0)
         )
-        if meta_curriculum_reset or generalization_reset:
+        if meta_curriculum_reset or generalization_reset or self.ghost_controller.reset:
             self.end_trainer_episodes(env, lessons_incremented)
 
     @timed

ml-agents/mlagents/trainers/trainer_util.py

         train_model: bool,
         load_model: bool,
         seed: int,
+        ghost_controller: GhostController,
         init_path: str = None,
         meta_curriculum: MetaCurriculum = None,
         multi_gpu: bool = False,
         self.seed = seed
         self.meta_curriculum = meta_curriculum
         self.multi_gpu = multi_gpu
-        self.ghost_controller = GhostController()
+        self.ghost_controller = ghost_controller
 
     def generate(self, brain_name: str) -> Trainer:
         return initialize_trainer(

ml-agents/setup.py

         "six>=1.12.0",
         "tensorflow>=1.7,<3.0",
         'pypiwin32==223;platform_system=="Windows"',
+        # We don't actually need six, but tensorflow does, and pip seems
+        # to get confused and install the wrong version.
+        "six>=1.12.0",
     ],
     python_requires=">=3.6.1",
     entry_points={

com.unity.ml-agents/Tests/Editor/TestModels.meta

+fileFormatVersion: 2
+guid: 95997790219c547e584c3cb50122a95f
+folderAsset: yes
+DefaultImporter:
+  externalObjects: {}
+  userData: 
+  assetBundleName: 
+  assetBundleVariant: 

utils/run_markdown_link_check.py

+#!/usr/bin/env python3
+
+import argparse
+import subprocess
+
+if __name__ == "__main__":
+    # markdown-link-check doesn't support multiple files on the commandline, so this hacks around that.
+    # Note that you must install the package separately via npm. For example:
+    #  brew install npm; npm install -g markdown-link-check
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--check-remote", action="store_true")
+    parser.add_argument("files", nargs="*")
+    args = parser.parse_args()
+
+    config_file = (
+        "markdown-link-check.full.json"
+        if args.check_remote
+        else "markdown-link-check.fast.json"
+    )
+
+    for f in args.files:
+        subprocess_args = ["markdown-link-check", "-c", config_file, f]
+        subprocess.check_call(subprocess_args)

com.unity.ml-agents/Tests/Editor/Resources.meta

-fileFormatVersion: 2
-guid: 22f1c3c8541da48e480c6b921343c2ee
-folderAsset: yes
-DefaultImporter:
-  externalObjects: {}
-  userData: 
-  assetBundleName: 
-  assetBundleVariant: